2023 International Conference on Computer Communication and Informatics (ICCCI ), Jan.

23 – 25, 2023, Coimbatore, INDIA

Shopify – Shop Augmentation and Recommendation

System
Subhojeet Bhattacharjee Nishant Singh N. Prasath
2023 International Conference on Computer Communication and Informatics (ICCCI) | 979-8-3503-4821-7/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICCCI56745.2023.10128209

Department of Networking and Department of Networking and Associate Professor - Department of

Communications Communications Networking and Communications
SRM Institute of Science and SRM Institute of Science and SRM Institute of Science and
Technology Technology Technology
Chennai, India Chennai, India Chennai, India
[email protected] [email protected] [email protected]

Abstract— Modern times have brought about changes in lifestyle. with a street vendor, it might be difficult to tell the difference
People dislike visiting crowded markets because they take too between an entrepreneur and his or her company, especially
long and are uncomfortable. Thus, an advantage of online when issues develop. The problem could be associated with a
shopping is, it saves tons of time. Online purchasing is a certain person, thing, event, or procedure. Some problems, or
procedure where clients go directly to a vendor on the Internet to problems inside the business unit, are brought on by factors
buy products, services, etc. Customers can browse online outside the control of the company, and other problems are
merchants from the convenience of their home and make created by factors inside the control of the corporation. The
purchases while seated at a computer. This research aims to first is easier to manage than the second [7]. The
introduce local people to the online shopping application so that
recommendation system, which proposes goods tailored to the
local merchants can profit from it and expand their reach and
sales through this fantastic platform. This research aims to give
consumer based on the preferences of the client, is one method
clients of physical stores the benefits of internet shopping. It of making shopping easier for customers [8]. Managing
facilitates purchasing goods from any store using a smart device massive amounts of data for consumers is one of the RSs'
and the internet. In this way, the customer will benefit from objectives. RS increases EC sales by bringing in new
home delivery and online shopping services from his preferred customers, showcasing new products, developing client
retailer. The solution is different in an umpteen number of ways. loyalty, raising customer happiness, and raising the likelihood
It is a direct exchange between the buyer and the vendor. This that a happy customer will come back. Studies show that RSs
system can be applied to any local business or to international are an essential component of the electronic industry, with
brands with retail store chains in the local areas. personalized RSs improving sales by up to 35% through
suggested items. As a result, these technologies are playing a
Keywords- Encryption, Preference Matrix, Feature-based bigger role as a way to do person-to-person marketing. [9]. One
product analysis, Chatbot, AOT Compilation, JPA, Scalability, of the most well-known and commonly used recommender
Security systems is the collaborative filter (CF), which is considered as
one of the key elements of effective E-commerce systems [10-
I. INTRODUCTION 11].
E-commerce has risen to prominence as one of the new The major challenges consist of:
purchasing trends on the internet. Online shopping has several
benefits, including time savings and the flexibility to buy goods 1. Boot-start: The RS works poorly when there is minimal
anywhere, anytime [1]. Any business dealing with the transfer data or metadata available. It is constantly accessible to both
of ownership or employment rights to goods or services new and regular users [12-13].
through a computer network is typically considered to be part 2. Customers' preference to simply rate a few things leads
of online buying and selling over the internet [2-4]. Three to data sparsity [14-15].
subcategories of e-commerce exist: business-to-business (B2B)
(Cisco), business-to-consumer (B2C) (Amazon), and 3. Scalability: Scalability issues have exploded as e-
consumer-to-consumer (C2C) (eBay). Developing e-commerce commerce websites have rapidly expanded [16].
websites involves both front-end and back-end development. 4. Diversity: Instead of focusing on differences, suggestions
Front-end development can make use of HTML, CSS, emphasis overlaps. The user is thereby exposed to fewer items
JavaScript, J-query, and ajax. The main industry norm for overall, and highly important specialized items could go
creating Web pages is the hypertext markup language (HTML). missed [17].
It mostly serves to describe the Web pages' element-based
structure [5-6]. Problems are frequently caused by procedural 5. One of the main challenges was selecting the project
hiccups, an honest perception of some difficulty, a failure to stack from a large selection of frameworks and tools [18].
keep promises, and the desire for perfection. When dealing

979-8-3503-4821-7/23/$31.00 ©2023 IEEE

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on May 31,2023 at 03:29:10 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Computer Communication and Informatics (ICCCI ), Jan. 23 – 25, 2023, Coimbatore, INDIA

II. RELATED WORKS recommendations provided by each system are then combined
Numerous difficulties with the online buying system and using a variety of fusion techniques. The resulting hybrid
recommendation system have been raised in prior work, which system can be tuned for the desired metric using this technique.
has been cited. Performance of e-commerce systems may be According to the evaluation metrics, the hybrid approach can
improved with the help of recommendation systems. As a help with concerns relating to sparsity and variety.
result, several research regarding creating diverse Mojtaba Salehi 2019 [24] based on user behaviour,
recommendation systems utilizing various methodologies were provided some helpful tips. A preference matrix was created to
conducted. Some of them were consumer behavior-focused and measure consumer preferences based on product qualities since
the studies are as follows: people communicate thoughts intuitively. Weighted association
Yajie Hu, Mitsunori Ogihara 2011 [19] proposed a fresh rules are also used to find these patterns, raising the standard of
strategy by gauging a song's novelty using the "Forgetting suggestions. The discovery of a user behavior-based
Curve" and measuring "preferences" using a user log data, the recommendation technique that determines users' preferences
system seeks to suggest songs that the user like, is unfamiliar to based on implicit product attribute ratings is the contribution
them, and fits their listening habits. To ascertain the user's that is most noteworthy. This approach also makes advantage
interest level in the next song and their listening preferences of a sequential buying pattern to raise the calibre of
are assessed. An analysis of this article based on the RS recommendations. This article ignores the remaining issues of
challenges reveals that when the cold-start issue cannot be the previous RSs and focuses on fixing sparsity and using the
fixed, a new client is given a song at random. There was an product's characteristics to assess how similar they are to one
increase in time complexity that required carrying out multiple another. Simply comparing two products to determine whether
computations due to the scalability of the number of songs and they are similar is insufficient to determine whether a
the vast diversity of customer tastes. recommendation is sound because not all customers have the
same preferences.
Bo Wang et al. in 2018 [20]. (BUIF) produced an implicit
feedback-based personalized recommendation algorithm. BUIF Duo Lin, Su Jingtao 2020 [25-26] developed a
considers the user's analysis and item sequencing in addition to groundbreaking recommender technique for extensive online
their purchasing behavior. The user's behavior log is used to shopping platforms. Products with higher preference degrees
extract the user's purchase, comparison, and item patterns. The are then offered to the client. The desired degree for each item
comparison and purchase behavior of the user are used to is chosen using contextual data from the client, such as access,
determine the user's similarity, and the word-to-product click, read, and purchase data. Because browsing, rating, and
embedding method is used to determine the item's similarity. searching were insufficiently important to include, this essay
Similarity measurements can be used to address the challenges mainly focused on the buy and click activities. Without
of variety and scalability, but not the others. addressing the other issues, it only managed to solve the
sparsity problem.
Kai Wang et al. 2019 [21] created a bespoke e-commerce
product recommendation system based on learning clustering All of the earlier work was mostly concentrated on static
representation. The ability of the conventional KNN technique modeling-oriented solutions, which were also less responsive
to choose close-by object sets is constrained. They provided a and reliable. With this research, it is possible to identify
way to pick the set of nearby items as well as a dynamic alternatives to the spring framework and angular that offer
selection model, time function, and neighborhood factor. The more security and are more durable for the long-term
e-commerce product suggestion system was developed by development of applications. No one today compromises on
combining RNN and attention processes. security, therefore this also looks at the security side of the
code and produces highly responsive web-based technology
Liao et al [22] proposed knowledge patterns, rules, and that is more swift, resilient, and secure.
knowledge maps were proposed, explained, and solutions for
online shopping were provided. Knowledge extraction through III. PROPOSED WORK
data mining. Kuo et al suggested the idea of post-purchase
intentions and post-recovery consumer pleasure. Additionally, To assist business owners and consumers in overcoming
negative emotions decrease post-recovery satisfaction whereas their difficulties, a user-friendly e-commerce website with a
positive emotions boost it. low cost has been developed that enables clients to select their
products and make a home buy. The merchants can provide
Andres Ferraro et al 2018's [23] focused on music every product through a single window in a manner similar to
recommendations and suggested a novel approach to enhance that.
suggestions based on a desired metric of choice that may be
achieved by combining different algorithms for each user A. The Description of the Proposed System
independently in accordance with their expected performance.
The suggested approach necessitates predicting an error that The idea of the research paper involves a good deal of
each system will commit for every user based on that user's deliberation over the idea of using multiple web-based
previous actions. For a number of measures that forecast technologies present at the current scenario and cloud
system performance based on a variety of characteristics that infrastructure. In general Shopify is an E-commerce product
describe historical user activity in the system, the study which tries to reduce the distance between a vendor and a
recommended a training regression model. The buyer. An ingenious attempt to alleviate the problems of a
buyer while preparing his shopping list. It offers the vendor

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on May 31,2023 at 03:29:10 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Computer Communication and Informatics (ICCCI ), Jan. 23 – 25, 2023, Coimbatore, INDIA

stores and their products at the fingertips of the customers. Every product in the product list has an empty queue:
What really is the key idea and belief that one day this will for every product in the queue
make the notion of "Products at our fingertips" true. Shopify if product_match == value
tries to improve the market base of vendors in their respective return the products's location
areas by increasing their visibility. It provides the ability to end if
buyers to filter out shops in their areas based on multiple end for
filtering algorithms. The two approaches to do effective ADD Product
resource management are: first, by leveraging elastic load SORT the products in following orders:
balancers and the cloud to deploy servers as needed, and Purchased products in descending order
second, by using the spring tool suite to improve the Viewed products in descending order
application's security. Better shared functionality can be Disliked products in ascending order
enabled by using the aforementioned methods in order to share Sort(array, left_pointer, right_pointer):
the same platform or web application without interfering with if left_pointer> right_pointer
other modules. Such functionality aids in system upkeep, return
internal cost control, or charging for third-party services. middle value = (left_pointer + right_pointer)/2
Additionally, this offers the web application sluggish loading. Sort(array, left_pointer, middle_value pointer)
Sort(array, middle_value+1, right_pointer)
This research was conducted using both primary and merge(array, left_pointer, mid_pointer
secondary sources. The following sections contained this , right_pointer)
collection of data. recommend_List = Merge all
Primary data were gathered from internet shoppers in town end
using a methodically created questionnaire. A questionnaire • Algorithm 2 for automated chatbot
containing the respondents' preferences as well as their age,
gender, and educational background was created in order to Input: User’s Natural language is taken as the input.
conduct statistical inquiries. Output: A natural language form and the
conversation log were used to request the
Secondary information was gathered from a variety of value of the MES data.
books, journals, research papers, and websites. Algorithm: User participation
Vendors can use their credentials to sell their goods by 1. while client ask a qustion
including a detailed description and images. Items include 2. client enters the query into queue
things like phones, laptops, computers, and their accessories, as 3. recognized, and keyword extraction
well as things like books, kitchen and home appliances, and so 4. for each keyword key in queue
on. In a private or public conversation, the seller can review 5. look for key in MES data set
advertising, control bidders, rate buyers, ask for a payout, and 6. save key for next loop
address questions from bidders. The seller determines the 7. return MES data value=val
sufficient price for the item rather than setting a minimum 8. end for
price. Additionally, the seller chooses the window of 9. val predetermined response string concatenation
opportunity during which bidders may submit their offers; 10. end while
during this window, the seller may decide to feel comfortable
selling the item. The buyer will be informed via email or text C. Architectural Diagram
message that they are not able to put a bid on the chosen item if The architecture design shown in Figure 1 was used to
the seller chooses to block bids. The seller can create a list of execute the proposed solution. Both the customer and the seller
pre-approved bids to ensure privacy while selling the goods to can log in or join up for the application as clients here. The
a chosen group of buyers. A trade broker who is an data will be checked against the database's existing data. The
experienced seller supports the inexperienced sellers in their data will be shown as the front-end using Angular, HTML,
product bidding as a guidance for bidders. CSS, and TypeScript after successful validation. To build a
web socket and handle Web Socket events and actions, the
Our study's goal is to make more prominent search options
front-end will need a web socket server. It offers methods
available and generate the most relevant results. The proposed
comparable to the Web Socket client API, and most
work is represented by Algorithm 1:
programming languages support it. The back-end server is built
with Spring Boot and Java, and it sends and receives data from
B. Algorithms Used and from the front-end. The database is used to hold all of the
• Algorithm 1 for shop augmentation and client and product records in separate tables. The server
recommendation retrieves this information via data transfer. Amazon Aurora, an
AWS service, is used to deploy the database to the cloud.
Input: Buyer id, Vendor id, Products_id, Customer
feedback
Output: Start the recommender queue
INITIALIZE:
Buyer_id = Vendor_id = Products_id = 0

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on May 31,2023 at 03:29:10 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Computer Communication and Informatics (ICCCI ), Jan. 23 – 25, 2023, Coimbatore, INDIA

Figure 3. Describes complete vendor module along with multiple services

Figure 1. The complete architectural flow of shop augmentation and
recommendation System
D. Database Design of Shopify
The system is made up of four tables which are built using
The vendor module was run using the architectural design MySQL relational database: one with basic information about
depicted in Figure 2. The customer will log in/sign up and the buyer, one with basic information about the product
then be taken to the dashboard. The customer may use the categories, one with basic information about the suppliers, and
dashboard to add, update, remove, and view products, among one with the order list. The connection between the linked
other things. Following these procedures, a preference matrix fields in the tables allows us to determine the relationship.
will be constructed, which will aid in the creation of the Additionally, Figure 4 illustrates the relationship between the
recommendation system. The operations are transmitted to the tables:
server, which retrieves the data from the database. A feedback
system will be maintained to identify the client base's
preferences in his/her searches and activities. The
recommendation system will create a list of suggestions that
will be utilized in the future.

Figure 2. Describes complete vendor module along with multiple services

The buyer module was executed using the architecture

concept depicted in Figure 3. The customer will log in/sign up
and then be taken to the dashboard. The customer may use the
dashboard to browse, search, filter, and evaluate items, among
other things. Following these procedures, a preference matrix
will be constructed, which will aid in the creation of the Figure 4. System Database Design
recommendation system. The operations are transmitted to the
server, which retrieves the data from the database. Based on
his or her previous searches, the buyer will be suggested IV. RESULTS DISCUSSIONS
preferred goods. The user may also search for products based RS determines how similar the users are based on how
on region, zip code, or product category. many common items they have purchased. RS uses the quantity
of common consumers who bought the same thing to assess
how comparable it is. listing of goods from various industries

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on May 31,2023 at 03:29:10 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Computer Communication and Informatics (ICCCI ), Jan. 23 – 25, 2023, Coimbatore, INDIA

that can be obtained from nearby stores serves as a common Figure 7 details the quantity of products in each category,
platform for both buyers and sellers when it comes to local allowing sellers to calculate their potential revenue and
shopping. These industries cover a variety of areas, such as determine how much of each product should be kept in store.
apparel, food products, raw materials, and groceries. Anyone
interested in selling products is welcome to sign up on Shopify.
Clustering coefficients for several graph classes are
displayed in Figure 5. Except for babies, all other categories
have a higher clustering coefficient than the corresponding
random graph (with the same number of edges).

Figure 7. Categories vs Number of items

The Ranking Technique was used to assess the data

that had been obtained using Preference Ranking Algorithm
below.
Figure 5. Categories vs Clustering Coefficient

Probability of a graph edge across many classes is depicted

in Figure 6. In terms of likelihood, the baby category is over
10% more likely than the grocery category. The likelihood is (1)
the sum of all potential edges in the undirected graph (n(n-1)), Using a matrix table, the resultant rank for each issue is
where n is the number of vertices) of products that were both converted into a score. The sum of each respondent's scores is
seen or purchased together. divided by the total number of respondents whose scores are
added together. The rankings are presented in ascending order
according to importance, with the mean scores for each factor.
The ranking approach is used to examine online purchasers'
preferences for products.
A. Discussions
The proposed work is able to declare that diversity and
scalability issues are under control because the suggested
website that sells computers and the things that go with them
uses the suggested RS (Recommended System) system.
Problems like these can be effectively addressed in more
comprehensive e-commerce settings including many
categories, such as apparel, electronics, and personal care
items, by employing additional factors categorized according to
the products' categories. Figure 6 depicts the probability of an
edge, while Figure 5 depicts the clustering coefficient (CC).
There is a strong relationship between these two metrics (R =
0.85). These two quantities are symbolic of the cohesiveness of
the product-to-product network that how often two products are
considered or purchased together. The grocery, kitchen, and
Figure 6. Categories vs Probability of an Edge toy categories appear to have a good level of clustering, but
when compared to the CC of all categories, there is space for
improvement.

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on May 31,2023 at 03:29:10 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Computer Communication and Informatics (ICCCI ), Jan. 23 – 25, 2023, Coimbatore, INDIA

V. CONCLUSION [11] Shareef, Sarah M., and Soukaena H. Hashim. "Proposed hybrid classifier
to improve network intrusion detection system using data mining
The In this study, a recommendation system is offered techniques." Engineering and Technology Journal 38, no. 1 (2020): 6-
along with a discussion of the RS (Recommended Systems) 14.
concerns. Scalability, diversity, sparsity, and cold start are a [12] Zhang, Hongwei, Xiangwei Kong, and Yujia Zhang. "Selective
few of these challenges. As seen in linked papers, this paper Knowledge Transfer for Cross-Domain Collaborative
Recommendation." IEEE Access 9 (2021): 48039-48051.
addresses some of these issues but not all of them. The
suggested system computes a number of characteristics [13] Wang, Donghui, Yanchun Liang, Dong Xu, Xiaoyue Feng, and Renchu
Guan. "A content-based recommender system for computer science
(consumer behavior) using statistical methods and analysis, and publications." Knowledge-Based Systems 157 (2018): 1-9.
then presents a list of recommendations that are customized to [14] Mohammed, Mazin Abed, Dheyaa Ahmed Ibrahim, and Akbal Omran
the users' tastes. Salman. "Adaptive intelligent learning approach based on visual anti-
spam email model for multi-natural language." Journal of Intelligent
Future work might be to analyze how the various categories Systems 30, no. 1 (2021): 774-792.
are combined. Observing how the products cluster when the [15] Sahoo, Abhaya Kumar, Chittaranjan Pradhan, Rabindra Kumar Barik,
graph is made up of pairs of categories may be especially and Harishchandra Dubey. "DeepReco: deep learning based health
useful. This would aid in identifying the product categories that recommender system using collaborative filtering." Computation 7, no.
a buyer is most likely to purchase and boost vendor revenue. 2 (2019): 25.
[16] Fayyaz, Zeshan, Mahsa Ebrahimian, Dina Nawara, Ahmed Ibrahim, and
Rasha Kashef. "Recommendation systems: Algorithms, challenges,
REFERENCES metrics, and business opportunities." applied sciences 10, no. 21 (2020):
7748.
[1] Liao, Shu-hsien, Pei-hui Chu, Yin-ju Chen, and Chia-Chen Chang. [17] Mostafa, Salama A., Saraswathy Shamini Gunasekaran, Aida Mustapha,
"Mining customer knowledge for exploring online group buying Mazin Abed Mohammed, and Wafaa Mustafa Abduallah. "Modelling an
behavior." Expert Systems with Applications 39, no. 3 (2012): 3708- adjustable autonomous multi-agent internet of things system for elderly
3716. smart home." In International conference on applied human factors and
ergonomics, pp. 301-311. Springer, Cham, 2019.
[2] Niranjanamurthy, M., N. Kavyashree, S. Jagannath, and Dharmendra
Chahar. "Analysis of e-commerce and m-commerce: advantages, [18] Kuanr, Madhusree, B. Kesari Rath, and S. Nandan Mohanty. "Crop
limitations and security issues." International Journal of Advanced recommender system for the farmers using mamdani fuzzy inference
Research in Computer and Communication Engineering 2, no. 6 (2013): model." International Journal of Engineering & Technology 7, no. 4.15
2360-2370. (2018): 277-280.
[3] Sivaji, Ashok, Alan G. Downe, Muhammad Fahmi Mazlan, Shi-Tzuaan [19] Sulikowski, Piotr, and Tomasz Zdziebko. "Deep learning-enhanced
Soo, and Azween Abdullah. "Importance of incorporating fundamental framework for performance evaluation of a recommending interface
usability with social & trust elements for e-commerce website." In 2011 with varied recommendation position and intensity based on eye-
International Conference on Business, Engineering and Industrial tracking equipment data processing." Electronics 9, no. 2 (2020): 266.
Applications, pp. 221-226. IEEE, 2011. [20] Srifi, Mehdi, Ahmed Oussous, Ayoub Ait Lahcen, and Salma Mouline.
[4] Gupta, Swati. "Online shopping cart application." (2013). "Recommender systems based on collaborative filtering using review
texts—a survey." Information 11, no. 6 (2020): 317.
[5] Buzzi, Maria Claudia, Marina Buzzi, Barbara Leporini, and Fahim
Akhter. "User trust in ecommerce services: perception via screen [21] Wei-wei, Guo, and Liu Feng. "Application Research of Hadoop's Weibo
reader." In 2009 international conference on new trends in information Recommendation System Prototype Based on Customer Dynamic
and service science, pp. 1166-1171. IEEE, 2009. Behavior." In 2018 3rd International Conference on Smart City and
Systems Engineering (ICSCSE), pp. 562-566. IEEE, 2018.
[6] Kiran, PB Narendra, and G. N. P. V. Babu. "Problems and prospects of
street vendors: a study with reference to Visakhapatnam city." [22] Park, Chanyoung, Donghyun Kim, Min-Chul Yang, Jung-Tae Lee, and
International Journal of Management, Technology and Engineering 9, Hwanjo Yu. "Click-aware purchase prediction with push at the top."
no. 6 (2019): 2249-7455. Information Sciences 521 (2020): 350-364.
[7] Elahi, Ehtsham, and Ashok Chandrashekar. "Learning representations of [23] Li, Xixi, Jiahao Xing, Haihui Wang, Lingfang Zheng, Suling Jia, and
hierarchical slates in collaborative filtering." In Fourteenth ACM Qiang Wang. "A hybrid recommendation method based on feature for
Conference on Recommender Systems, pp. 703-707. 2020. offline book personalization." arXiv preprint arXiv:1804.11335 (2018).
[8] Abbas, Ayad R. "An Adaptive E-Commerce: Applying of Psychological [24] Jiao, Ming-hai, Xiao-fang Chen, Zhong-hai Su, and Xu Chen. "Research
Testing Method to Improve Buying Decision Process." Engineering and on personalized recommendation optimization of E-commerce system
Technology Journal 33, no. 2 Part (B) Scientific (2015). based on customer trade behaviour data." In 2016 Chinese Control and
Decision Conference (CCDC), pp. 6506-6511. IEEE, 2016.
[9] Bortko, Kamil, Piotr Bartków, Jarosław Jankowski, Damian Kuras, and
Piotr Sulikowski. "Multi-criteria evaluation of recommending interfaces [25] Li, Dancheng, Guangming Zhao, Zhi Wang, Wenjia Ma, and Ying Liu.
towards habituation reduction and limited negative impact on user "A method of purchase prediction based on user behavior log." In 2015
experience." Procedia Computer Science 159 (2019): 2240-2248. IEEE International Conference on Data Mining Workshop (ICDMW),
pp. 1031-1039. IEEE, 2015.
[10] Lee, Dokyun, and Kartik Hosanagar. "How do recommender systems
affect sales diversity? A cross-category investigation via randomized [26] Xu, Jing, Jie Wang, Ye Tian, Jiangpeng Yan, Xiu Li, and Xin Gao. "SE-
field experiment." Information Systems Research 30, no. 1 (2019): 239- stacking: Improving user purchase behavior prediction by information
259. fusion and ensemble learning." Plos one 15, no. 11 (2020): e0242629.

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on May 31,2023 at 03:29:10 UTC from IEEE Xplore. Restrictions apply.